CODEWITHSUNDEEP
machine-learninglibsvmsvm
Marc L'Heureux
Marc L'Heureux

Reputation: 412

Support Vector Machine distance from hyperplane and certainty

Does a vector's distance from the decision plane in a support vector machine (with arbitrary configuration with regards to kernel, etc.) correlate in any way to the machine's confidence/certainty of that vector's classification?

Upvotes: 2

Views: 1409

Answers (2)

Rob Neuhaus
Rob Neuhaus

Reputation: 9290

Yes. This is why Platt Scaling, which monotonically transforms the margin can produce calibrated probabilities, for example.

However, without transforming the margin, it's hard to give a workable interpretation.

Upvotes: 3

Alptigin Jalayr
Alptigin Jalayr

Reputation: 709

Yes, I would most certainly say it does. Some formalization given here: http://cs229.stanford.edu/notes/cs229-notes3.pdf

Upvotes: 1

Related Questions